gábor c. temes - hasler.ece.gatech.edu · [email protected] cascade filter design [3], [5] •...

of 99/99
[email protected] 1/99 Gábor C. Temes School of Electrical Engineering and Computer Science Oregon State University

Post on 20-Jun-2020

0 views

Category:

Documents

0 download

Embed Size (px)

TRANSCRIPT

  • [email protected] 1/99

    Gábor C. Temes

    School of Electrical Engineering and Computer Science

    Oregon State University

  • [email protected]

    Mixed-Mode Electronic System

    •  Analog filters needed to suppress out-of-band noise and prevent aliasing. Also used, e.g., as loop filters in oversampled ADCs.

    •  Continuous-time filter allows sampling by discrete-time SCF. SCF performs sharper filtering; DSP filtering may be even sharper.

    2/99

  • [email protected]

    Frequency Range of Analog Filters

    •  Discrete active RC filters: 1 Hz – 100 MHz

    •  On-chip continuous-time active filters: 10 Hz - 1 GHz

    •  Switched-capacitor or switched-current filters: 1 Hz – 10 MHz

    •  Discrete LC: 10 Hz - 1 GHz

    •  Distributed: 100 MHz – 100 GHz

    3/99

  • [email protected]

    Accuracy Considerations

    •  The absolute accuracy of on-chip analog components is bad (10% - 50%). The matching accuracy of like elements is much better.

    •  In analog integrated circuits, on-chip Rs can be matched to each other typically within a few %, Cs within 0.05%, with careful layout. Transconductance (Gm) stages can be matched to about 10 - 30%.

    •  In an active RC filter, the time constant Tc is determined by RC products, accurate to only 20 – 50%. In a Gm-C filter , Tc ~ C/Gm, also inaccurate. Tuning may be used.

    •  In an SC filter, Tc ~ (C1/C2)/fc, where fc is the clock frequency. Tc accuracy may be 0.05% or better!

    4/99

  • [email protected]

    Classification of Filters •  Signal types: continuous-time vs. discrete-time filters;

    •  CT: signal is a continuous function of time. Laplace and Fourier transforms can be used for analysis. Example: Gm-C filters, SCFs in Sit.2 (slide 2).

    •  DT: only samples processed. Z-transform and DFT must be used. Example: SCFs in Sit.1 (slide 2).

    •  Signal representation: voltage, charge, current.

    •  Filter architectures: cascade vs. multi-feedback structures.

    5/99

  • [email protected]

    Active-RC Filters [1], [4], [5]

    •  Single-amplifier filters: Sallen-Key filter; Kerwin filter; Rauch filter, Deliyannis-Friend filter.

    •  Integrator-based filter sections: Tow-Thomas biquads; Ackerberg-Mossberg filter.

    •  Cascade design, dynamic range optimization.

    •  Simulated LC filters: gyrator-based and integrator-based filters; dynamic range optimization.

    6/99

  • [email protected]

    Sallen-Key Filter [1],[4]

    Single-opamp biquad. General diagram:

    Often, K = 1. 5 para- meters, 3 specified values.

    •  Amplifier not grounded. Common-mode changes with output. Differential implementation difficult.

    7/99

  • [email protected]

    Sallen-Key Filter

    •  Transfer function:

    •  Second-order transfer function (biquad) if two of the admittances are capacitive. Complex poles achieved

    by subtraction of term containing K.

    8/99

    H(s) = V2V1

    =KY1Y3

    (Y1 +Y2)(Y3 +Y4 ) +Y3Y4 −KY2Y3

  • [email protected]

    Sallen-Key Filter

    •  Low-pass S-K filter (R1, C2, R3, C4):

    •  Highpass S-K filter (C1, R2, C3, R4):

    •  Bandpass S-K filter ( R1, C2, C3, R4 or C1, R2, R3, C4):

    9/99

  • [email protected]

    Sallen-Key Filter •  Pole frequency ωo: absolute value of natural mode; •  Pole Q: ωo/[2|real part of pole|]. Determines the

    stability, sensitivity, and noise gain. Q > 5 is dangerous, Q > 10 can be lethal! For S-K filter

    •  dQ/Q ~ (3Q –1) dK/K .So, if Q = 10, 1% error in K results in 30% error in Q.

    •  Pole Q tends to be high in band-pass filters so S-K may not be suitable for those.

    •  Usually only the peak gain, the Q and the pole frequency ωo are specified. There are 2 extra degrees of freedom. May be used for specified R noise, minimum total C, equal capacitors, or K = 1.

    •  Use differential difference amplifier for differential circuit.

    10/99

  • [email protected]

    Kerwin Filter •  Sallen-Key filters cannot realize finite imaginary zeros,

    needed for elliptic or inverse Chebyshev response. Kerwin filter can with Y = G or sC.

    Y = G:

    Y = saC:

    11/99

    H(s) = K(s2 + a−2)

    s2 + (G + 2 −K)2a−1s+ (1+ 2G)a−2

    H(s) = [K /(2C +1)](s2 + a−2)

    s2 + [2a−1(C + 2 −K) /(2C +1)]s+ a−2(2C +1)

  • [email protected]

    Deliyannis-Friend Filter [1] Single-opamp bandpass filter:

    •  Grounded opamp, Vcm = 0. The circuit may be realized in a fully differential form suitable for noise cancellation. CM is analog ground.

    •  Finite gain slightly reduces gain factor and Q. Sensitivity is not too high even for high Q. Q Q - K/A.

    12/99

  • [email protected]

    Deliyannis-Friend Filter [1]

    For finite opamp gain:

    For infinite opamp gain:

    13/99

    H(s) = V2V1

    =N(s)D(s)

    = −A

    1+ A⎛

    ⎝ ⎜

    ⎠ ⎟

    saG1 /C

    s2 + s 2G2C

    +G1 /C1+ A

    ⎝ ⎜

    ⎠ ⎟ +

    G1G2C2

    H(s) = − saCG1s2C2 + 2sCG2 +G1G2

    = −saG1 /C

    s2 + s2G2 /C +G1G2 /C2

  • [email protected]

    Deliyannis-Friend Filter

    •  Q may be enhanced using positive feedback:

    •  New Q =

    •  Opamp no longer grounded, Vcm not zero, no easy fully differential realization.

    14/99

  • [email protected]

    Rauch Filter [1]

    •  Often applied as anti-aliasing low-pass filter:

    •  Transfer function:

    •  Grounded opamp, may be realized fully differentially. 15/99

  • [email protected]

    Tow-Thomas Biquad [1]

    •  Multi-opamp integrator-based biquads: lower sensitivities, better stability, and more versatile.

    •  They can be realized in fully differential form. •  The Tow-Thomas biquad is a sine-wave oscillator,

    stabilized by one or more additional element:

    •  In its differential form, the inverter is not required.

    16/99

  • [email protected]

    Tow-Thomas Biquad

    •  The transfer functions to the opamp outputs are

    Other forms of damping are also possible. Preferable to damp the second stage, or (for high Q) to use a damping C across the feedback resistor. Zeros can be added by feed-forward input branches.

    17/99

  • [email protected]

    Ackerberg–Mossberg Filter [1]

    •  Similar to the Tow-Thomas biquad, but less sensitive to finite opamp gain effects.

    •  The inverter is not needed for fully differential realization. Then it becomes the Tow-Thomas structure.

    18/99

  • [email protected]

    Cascade Filter Design [3], [5]

    •  Higher-order filter can constructed by cascading low-order ones. The Hi(s) are multiplied, provided the stage outputs are buffered.

    •  The Hi(s) can be obtained from the overall H(s) by factoring the numerator and denominator, and assigning conjugate zeros and poles to each biquad.

    •  Sharp peaks and dips in |H(f)| cause noise spurs in the output. So, dominant poles should be paired with the nearest zeros.

    19/99

  • [email protected]

    Cascade Filter Design [5] •  Ordering of sections in a cascade filter dictated by

    low noise and overload avoidance. Some rules of thumb:

    •  High-Q sections should be in the middle;

    •  First sections should be low-pass or band-pass, to suppress incoming high-frequency noise;

    •  All-pass sections should be near the input;

    •  Last stages should be high-pass or band-pass to avoid output dc offset.

    20/99

  • [email protected]

    Dynamic Range Optimization [3] •  Scaling for dynamic range optimization important in multi-

    opamp filters. •  Active-RC structure:

    •  Op-amp output swing must remain in linear range, but as large as this permits to reduce noise gain from the stage to the filter output.

    21/99

  • [email protected]

    Dynamic Range Optimization

    •  Multiplying all impedances connected to the opamp output by k, the output voltage Vout becomes k.Vout, and all output currents remain unchanged.

    •  Choose k.Vout so that the maximum swing occupies the linear range of the opamp.

    •  Find the maximum swing in the time domain by plotting the histogram of Vout for a typical input, or in the frequency domain by sweeping the frequency of an input sine-wave to the filter, and compare Vout with the maximum swing of the output opamp.

    22/99

  • [email protected]

    Optimization in Frequency Domain

    23/99

  • [email protected]

    Impedance Level Scaling •  Lower impedance -> lower noise, more power needed! •  All admittances connected to the input node of the opamp

    may be multiplied by a convenient scale factor without changing the output voltage or output currents. This may be used, e.g., to minimize the area of capacitors.

    •  Impedance scaling should be done after dynamic range scaling!

    24/99

  • [email protected]

    Tunable Active-RC Filters [2], [3]

    •  Tolerance of RC time constants typically 30 ~ 50%, so the frequency response may not be acceptable.

    •  Rs may be trimmed, or made variable and automatically tuned to obtain a time constant locked to the period T of a crystal-controlled clock signal.

    •  Simplest: replace Rs by MOSFETs operating in their linear (triode) region. MOSFET-C filters result.

    •  Compared to Gm-C filters, slower and need more power, but may be more linear, and easier to design.

    25/99

  • [email protected]

    Two-Transistor Integrators

    26/99

    •  Vc is the control voltage for the MOSFET resistors.

  • [email protected]

    Two-Transistor Integrators

    27/99

  • [email protected]

    MOSFET-C Biquad Filter [2], [3]

    28/99

    •  Tow-Thomas biquad:

  • [email protected]

    Four-Transistor Integrator

    29/99

    •  Improved linearity of MOSFET-C integrators by using 4 transistors rather than 2:

    •  May be analyzed by treating it as a two-input integrator with inputs (Vpi-Vni) and (Vni-Vpi).

  • [email protected]

    Four-Transistor Integrator•  If all four transistor are matched in size,

    30/99

    •  Model for drain-source current shows non-linear terms not dependent on controlling gate-voltage;

    •  All even and odd distortion products will cancel; •  Model only valid for older long-channel length

    technologies; •  In practice, about a 10 dB linearity improvement.

  • [email protected]

    Tuning of Active-RC Filters

    •  Rs may be automatically tuned to match to an accurate off-chip resistor, or to obtain an accurate time constant locked to the period T of a crystal-controlled clock signal:

    •  In equilibrium, R.C = T. Match Rs and Cs to the ones in the tuning stage using careful layout. Residual error 1-2%.

    31/99

  • [email protected]

    Switched-R Filters [6]

    •  Replace tuned resistors by a combination of two resistors and a periodically opened/closed switch.

    •  Automatically tune the duty cycle of the switch:

    32/99

  • [email protected]

    Simulated LC Filters [3], [5]

    •  A doubly-terminated LC filter with near-optimum power transmission in its passband has low sensitivities to all L & C variations, since the output signal can only decrease if a parameter is changed from its nominal value.

    33/99

  • [email protected]

    Simulated LC Filters

    •  Simplest: replace all inductors by gyrator-C stages:

    •  Using transconductances:

    34/99

  • [email protected]

    Simulated LC Filters Using Integrators

    35/99

  • [email protected]

    Cascade vs. LC Simulation Design

    •  Cascade design: modular, easy to design, lay out, trouble-shoot. Passband sensitivities moderate (~0.3 dB), stopband sensitivities excellent, since stopband losses of the sections add.

    36/99

  • [email protected]

    Cascade vs. LC Simulation Design

    •  LC simulation: passband sensitivities (and hence noise suppression) excellent due to Orchard’s Rule.

    •  Stopband sensitivities high, since suppression is only achieved by cancellation of large signals at the output:

    37/99

  • [email protected]

    Gm-C Filters [1], [2], [5]

    •  Alternative realization of tunable continuous-time filters: Gm-C filters.

    •  Faster than active-RC filters, since they use open-loop stages, and (usually) no opamps..

    •  Lower power, since the active blocks drive only capacitive loads.

    •  More difficult to achieve linear operation (no feedback).

    38/99

  • [email protected]

    Gm-C Integrator

    •  Uses a transconductor to realize an integrator; •  Output current of Gm is (ideally) linearly related to input

    voltage; •  Output impedance is ideally infinite. •  Gm is not an operational transconductance amplifier

    (OTA) which has a high Gm value but is not very linear.

    39/99

  • [email protected]

    Multiple-Input Gm-C Integrator

    40/99

    •  Can process several inputs:

  • [email protected]

    Fully-Differential Integrators•  Better noise and linearity than for single-ended operation:

    •  Uses a single capacitor between differential outputs. •  Requires some sort of common-mode feedback to set

    output common-mode voltage. •  Needs some extra capacitors for compensating the

    common-mode feedback loop.

    41/99

  • [email protected]

    Fully-Differential Integrators

    •  Uses two grounded capacitors; •  Still requires common-mode feedback but here the

    compensation capacitors for the common-mode feedback can be the same grounded capacitors as used for the signal.

    42/99

  • [email protected]

    Fully-Differential Integrators

    •  Integrated capacitors have top and bottom plate parasitic capacitances.

    •  To maintain symmetry, usually two parallel capacitors turned around are used, as shown above.

    •  Note that parasitic capacitance affects the time constant.

    43/99

  • [email protected]

    Gm-C-Opamp Integrator

    •  Uses an extra opamp to improve linearity and noise performance.

    •  Output is now buffered.

    44/99

  • [email protected]

    Gm-C-Opamp IntegratorAdvantages •  Effect of parasitics reduced by opamp gain —more

    accurate time constant and better linearity. •  Less sensitive to noise pick-up, since transconductor

    output is low impedance (due to opamp feedback). •  Gm cell drives virtual ground — output impedance of Gm

    cell can be lower, and smaller voltage swing is needed. Disadvantages •  Lower operating speed because it now relies on feedback; •  Larger power dissipation; •  Larger silicon area.

    45/99

  • [email protected]

    A Simple Gm-C Opamp Integrator

    •  A low input impedance on the order of due to common-gate input impedance and feedback.

    46/99

    • Pseudo-differential operation. Simple opamp:

  • [email protected]

    First-Order Gm-C Filter

    •  General first-order transfer-function

    •  Built with a single integrator and two feed-in branches. •  ω0 sets the pole frequency

    47/99

  • [email protected]

    First-Order Filter

    48/99

    •  Transfer function is given by

  • [email protected]

    Fully-Differential First-Order Filter

    •  Same equations as for single-ended case, but capacitor sizes are doubled.

    •  3 coefficients, 4 parameters. May make Gm1 = Gm2. •  Can also realize K1 < 0 by cross-coupling wires at Cx.

    49/99

  • [email protected]

    Second-Order Filter

    50/99

    •  Tow-Thomas biquad:

  • [email protected]

    Second-Order Filter

    51/99

    •  Fully differential realization:

  • [email protected]

    Second-Order Filter

    •  There is a restriction on the high-frequency gain coefficients k2, just as in the first-order case (not for differential realization).

    •  Gm3 sets the damping of the biquad. •  Gm1 and Gm2 form two integrators, with unity-gain

    frequencies of ω0/s. 52/99

    • Transfer function:

  • [email protected]

    Second-Order Filter •  5 coefficients to match in H(s), 8 designable parameters

    (5 Gms, 3 capacitances).

    •  Extra degrees of freedom may be used for dynamic range at internal node and impedance scaling, and for using matched Gm blocks.

    •  In cascade design, the input admittance Yin is important. If Cx = 0, Yin = 0. Otherwise, it is Yin = sCx [ 1 – H(s)] .

    •  Yin may be absorbed in the previous stage’s output capacitor CB.

    53/99

  • [email protected]

    Scaling of Cascade Gm-C Filter •  In a cascade of biquads, H(s) = H1(s).H2(s). …. Before

    realization, scale all Hi(s) so that the maximum output swings are the largest allowable. This takes care of the output swings of Gm2, Gm3, and Gm5.

    •  Multiply Gm1 and Gm4, or divide CA, by the desired voltage scale factor for the internal capacitor CA. This takes care of the output swings of Gm1 and GM4.

    •  It is possible to multiply the Gms and capacitors of both integrators by any constant, to scale the impedances of the circuit at a convenient level (noise vs. chip area and power).

    54/99

  • [email protected]

    References [1] R. Schaumann et al., Design of Analog Filters (2nd

    edition), Oxford University Press, 2010. [2] D. A. Johns and K. Martin, Analog Integrated

    Circuits, Wiley, 1997. [3] R. Gregorian and G. C. Temes, Analog MOS

    Integrated Circuits for Signal Processing, Wiley, 1986.

    [4] Introduction to Circuit Synthesis and Design, G. C. Temes and J. W. LaPatra, McGraw-Hill, 1977.

    [5] John Khoury, Integrated Continuous-Time Filters, Unpublished Lecture Notes, EPFL, 1998.

    [6] P. Kurahashi et al., “A 0.6-V Highly Linear Switched-R-MOSFET-C Filter, CICC, Sept. 2006, pp. 833-836.

    55/99

  • [email protected]

    Switched-Capacitor Filters

    56/99

    Gábor C. Temes

    School of Electrical Engineering and Computer Science

    Oregon State University

  • [email protected] 57/99

    Switched-Capacitor Circuit Techniques [2], [3] •  Signal entered and read out as voltages, but processed

    internally as charges on capacitors. Since CMOS reserves charges well, high SNR and linearity possible.

    •  Replaces absolute accuracy of R & C (10-30%) with matching accuracy of C (0.05-0.2%);

    •  Can realize accurate and tunable large RC time constants;

    •  Can realize high-order dynamic range circuits with high dynamic range;

    •  Allows medium-accuracy data conversion without trimming;

    •  Can realize large mixed-mode systems for telephony, audio, aerospace, physics etc. Applications on a single CMOS chip.

    •  Tilted the MOS VS. BJT contest decisively.

  • [email protected] 58/99

    Competing Techniques

    •  Switched-current circuitry: Can be simpler and faster, but achieves lower dynamic range & much more THD; Needs more power. Can use basic digital technology; now SC can too!

    •  Continuous-time filters: much faster, less linear, less accurate, lower dynamic range. Need tuning.

  • [email protected] 59/99

    LCR Filters to Active-RC Filters

  • [email protected] 60/99

    Typical Applications of SC Technology – (1)

    Line-Powered Systems:

    •  Telecom systems (telephone, radio, video, audio) •  Digital/analog interfaces •  Smart sensors •  Instrumentation •  Neural nets. •  Music synthesizer

  • [email protected] 61/99

    Typical Applications of SC Technology – (2)

    Battery-Powered Micropower Systems:

    •  Watches •  Calculators •  Hearing aids •  Pagers •  Implantable medical devices •  Portable instruments, sensors •  Nuclear array sensors (micropower, may not be battery

    powered)

  • [email protected] 62/99

    New SC Circuit Techniques

    To improve accuracy: •  Oversampling, noise shaping •  Dynamic matching •  Digital correction •  Self-calibration •  Offset/gain compensation

    To improve speed, selectivity: •  GaAs technology •  BiCMOS technology •  N-path, multirate circuits

  • [email protected] 63/99

    Typical SC Stages

    •  Amplifiers: Programmable, precision, AGC, buffer, driver, sense

    •  Filters •  S/H and T/H stages •  MUX and deMUX stages •  PLLs •  VCOs •  Modulators, demodulators •  Precision comparators •  Attenuators •  ADC/DAC blocks

  • [email protected] 64/99

    Active - RC Integrator

    Can be transformed by replacing R1 by an SC branch.

  • [email protected] 65/99

    SC Integrator (Analog Accumulator) Stray insensitive version:

  • [email protected] 66/99

    SC Integrator (Analog Accumulator)

  • [email protected] 67/99

    Low-Q SC Biquad – (1)

  • [email protected] 68/99

    Low-Q SC Biquad – (2)

  • [email protected] 69/99

    Low-Q SC Biquad – (3) Approximate design equations for ω0T 1,

  • [email protected] 70/99

    Low-Q SC Biquad – (4) For b0 = 1, matching coefficients:

    Preliminary Ci values. Scaling to follow!

  • [email protected] 71/99

    High-Q Biquad – (1)

    (a) High Q switched-capacitor biquad; (b) clock and signal waveform

  • [email protected] 72/99

    High-Q Biquad – (2)

    Approximate design equations :

  • [email protected] 73/99

    High-Q Biquad – (3) Exact equations:

    For b2 = 1, coefficient matching gives C spread & sensitivities reasonable even for high Q & fc/fo, since C2, C3, C4 enter only in products:

  • [email protected] 74/99

    Linear Section – (1)

    Original φi : pole/zero btw 0 & 1.

    Parenthesized: zero > 1.

  • [email protected] 75/99

    Linear Section – (2)

    For fully differential circuit, less restrictions.

    Stray insensitive branch shown below (cannot be used as feedback branch!).

  • [email protected] 76/99

    Cascade Design – (1)

  • [email protected] 77/99

    Cascade Design – (2)

    Easy to design, layout, test, debug, Passband sensitivities “moderate,” Sens = 0.1 - 0.3 dB/% in passband. Stopband sensitivities good. Pairing of num. & denom., ordering of sections all affect S/N, element spread and sensitivities.

  • [email protected] 78/99

    Cascade Design – (3) Cascade Order

    Many factors practical factors influence the optimum ordering. A few examples [5]:

    1. Order the cascade to equalize signal swing as much as possible for dynamic range considerations

    3. Choose the first biquad to be a lowpass or bandpass to reject high frequency noise, to prevent overload in the remaining stages.

    3. lf the offset at the filter output is critical, the last stage should be a highpass or bandpass to reject the DC of previous stages

  • [email protected] 79/99

    Cascade Design – (4) 4.  The last stage should NOT in general be high Q because these

    stages tend to have higher fundamental noise and worse sensitivity to power supply noise

    6.  In general do not place allpass stages at the end of the cascade because these have wideband noise. It is usually best to place allpass stages towards the beginning of the filter.

    8.  If several highpass or bandpass stages are available one can place them at the beginning, middle and end of the filter. This will prevent input offset from overloading the filter, will prevent internal offsets of the filter itself from accumulating (and hence decreasing available signal swing) and will provide a filter output with low offset.

    7. The effect of thermal noise at the filter output varies with ordering; therefore, by reordering several dB of SNR can often be gained.

    (John Khouri, unpublished notes)

  • [email protected] 80/99

    Ladder Filter For optimum passband matching, for nominal ∂Vo/∂x ~ 0 since Vo is maximum x values. x: any L or C.

    Use doubly-terminated LCR filter prototype, with 0 flat passband loss. State equations:

  • [email protected] 81/99

    The Exact Design of SC Ladder Filters Purpose: Ha(sa) ↔ H(z), where

    Then, gain response is only frequency warped.

    State equations for V1,I1 &V3; Example:

    Purpose of splitting C1:

    has a simple z-domain realization.

  • [email protected] 82/99

    Sa-domain block diagram

    Realization of input branch: Qin=Vin/saRs, which becomes,

    This relation can be rewritten in the form

    or, in the time domain

    Δqin(tn) : incremental charge flow during tn-1 < t < tn, in SCF.

  • [email protected] 83/99

    Final Circuit

  • [email protected] 84/99

    Damping resistors in input & output stages – (2)

    Using bandpass realization tables to obtain low-pass response gives an extra op-amp, which can be eliminated:

    Sixth-order bandpass filter: LCR prototype and SC realization.

  • [email protected] 85/99

    To modify Vo → kVo ; Yi/Yf → kYi/Yf ∀i. Hence, change Yf to Yf /k or Yi to kYi. (It doesn't matter which; area scaling makes the results the same.) To keep all output currents unchanged, also Ya → Ya/k, etc.

    Noise gain :

    Scaling for Optimal DR and Chip Area – (1)

  • [email protected] 86/99

    Hence, . The output noise currents are also divided

    by k, due to Y′a = Ya/k, etc. Hence, the overall output noise from this stage changes by a factor

    where : the signal gain.

    Scaling for Optimal DR and Chip Area – (2)

  • [email protected] 87/99

    The output signal does not change, so the SNR improves with increasing k. However, the noise reduction is slower than 1/k, and also this noise is only one of the terms in the output noise power.

    If Vo> VDD, distortion occurs, hence k ≤ kmax is limited such that Yo saturates for the same Vin as the overall Vout. Any k > kmax forces the input signal to be reduced by k so the SNR will now decrease with k.

    Scaling for Optimal DR and Chip Area – (3)

    Conclusion: kmax is optimum.

  • [email protected] 88/99

    Purposes :

    1. Maximum dynamic range 2. Minimum Cmax / Cmin , ∑C / Cmin3. Minimum sens. to op-amp dc gain

    effects.

    Scaling of SCF's. – (1)

    1. Dynamic range Assume that opamps have same input noise v2n, and max. linear range |Vmax|. For an optimum dynamic range Vin max / Vin min, each opamp should have the same Vimax(f), so they all saturate at the same Vin max. Otherwise, the S/N of the op-amp is not optimal.

  • [email protected] 89/99

    To achieve V1 max = V2 max = ••• = Vout max, use amplitude scaling:

    Scaling of SCF's. – (2)

  • [email protected] 90/99

    Simple rule:

    Multiply all Cj connected or switched to the output of opamp i by ki!

    Scaling of SCF's. – (3)

  • [email protected] 91/99

    Scaling of SCF's. – (4)

    2. Minimum Cmax/Cmin: If all fn(z) & h(z) are multiplied by the same li, nothing will not change. Choose li = Cmin / Ci min where Ci min is the smallest C connected to the input of op-amp i, and Cmin is the smallest value of cap. permitted by the technology (usually 0.1 pF ≤ Cmin ≤ 0.5 pF for stray-insensitive circuits). Big effect on Cmax / Cmin!

    3. Sensitivities:

    The sensitivity of the gain to Ck remain unchanged by scaling; the sens. To op-amp gain effects are very much affected. Optimum dynamic-range scaling is nearly optimal for dc gain sens. as well.

  • [email protected] 92/99

    Scaling of SCF's. – (5)

  • [email protected] 93/99

    Scaling of SCF's. – (6) G

    AIN

    / dB

    FREQUENCY / Hz

  • [email protected] 94/99

    Scaling of SCF's. – (7) G

    AIN

    / dB

    FREQUENCY / Hz

  • [email protected] 95/99

    Scaling of SCF's. – (8) G

    AIN

    / dB

    FREQUENCY / Hz

  • [email protected] 96/99

    SC Filters in Mixed-Mode System Two situations; example:

    Situation 1: Only the sampled values of the output waveform matter; the output spectrum may be limited by the DSP, and hence VRMS,n reduced.

    Situation 2: The complete output waveform affects the SNR, including the S/H and direct noise components. Usually the S/H dominates. Reduced by the reconstruction filter.

  • [email protected] 97/99

    Direct-Charge-Transfer Stage – (1)

    Advantages: Opamp does not participate in charge transfer → no slewing distortion, clean S/H output waveform. Finite DC gain A, introduces only a scale factor K = 1/[1+1/Ao].

  • [email protected] 98/99

    Direct-Charge-Transfer Stage – (2) Analysis gives

    where

    is the ideal lowpass filter response.

    Applications:

    • SC-to-CT buffer in smoothing filter for D-S DAC (Sooch et al., AES Conv., Oct. 1991)

    • DAC + FIR filter + IIR filter (Fujimori et al., JSSC, Aug. 2000).

  • [email protected] 99/99

    Double Sampled Data Converter – (1)